標題: 社群網路上的潛在用戶探勘
Inferring Potential Users in Social Networks
作者: 徐宗豪
Hsu, Tsung-Hao
彭文志
網路工程研究所
關鍵字: 社群網路;潛在用戶;分類;社群;特徵選擇;social networks;potential users;classification;community;feature selection
公開日期: 2013
摘要: 隨著網路科技快速的發展,越來越多公司提供社群媒體的服務。對於服務提供者來說,越多的客戶使用他們提供的平台,他們將會有更高的營運收入。而如何找到潛在用戶並吸引他們加入服務,已經成為一項重要的議題。我們把有較高傾向加入某特定服務的使用者稱作”潛在用戶”。關於潛在用戶,我們能得到的資訊,只能藉由他們在某服務內的朋友來獲得間接的資訊。而在現實世界中,人們通常會受到朋友的影響。因此,藉由分析朋友互動的資訊,此篇論文利用一個間接的方式探勘出潛在用戶族群。我們先分析朋友間的互動資訊來抓取出一些explicit features。進一步地,因為人們經常會有屬於自己的社群(community),我們利用不同方式建立人與人之間的社群,並藉由這些社群抓取一些implicit features。為了找到更精確且有用的feature,我們進行了一系列的觀察,來找出effective feature set。有了上述的方法與觀察,我們利用分類器(classifier)來輔助我們預測潛在用戶。我們進行了綜合實驗在實際資料集上,結果顯示,我們的方法可以有效地預測潛在用戶,且達到接近70%的準確率。
With the developing of technologies about networks, there are more and more companies provide social media service. In service providers’ view, more customers lead to more income. How to explore new customers has become a significant issue. We call the people with high tendency to join a specific service as potential users. All the information about potential users comes from their friends. In the real world, people were often influenced by their friends. As a result, analyzing friends’ interaction behavior logs offer an unique way to explore potential users. In this paper, we extract explicit features based on friends’ interaction behavior. Moreover, people tend to organize their own community in their life, we extract community based implicit features for a deeper exploration. To select effective predictors, we do some observation for choosing discriminative feature set. After exploring the effective predictor, we use different classifiers to predict the potential users and compare their effectiveness. Finally, we conduct our method in real dataset and show that the features we extract can reach about 70% accuracy.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070056522
http://hdl.handle.net/11536/73114
Appears in Collections:Thesis


Files in This Item:

  1. 652201.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.